Just yesterday had a discussion with my colleagues about ‘alpine vs. centos base image’. By using alpine we can save memory for storing the image and also lower the volume of data flying in network. As a benefit final images atop the base image get distributed in the cluster a lot faster, which reduces application start time. If the image is not located on the node where the app gets scheduled on. But when alpine base image bites you in the butt is debugging. Just not having the most basic, standard tools makes life of an operator hell and as a result costs time & money.

Yes, centos image, with a proper bash and yum installer inside, is 196 MB, but container images (layers!) are reused. Therefore the base image is located exactly once at maximum on every node in cluster. Money spent on storage/network vs. money lost during system outage, where the ops guys struggle to debug the application, should be an easy decision to make.